If you write grants for a small team, someone has probably suggested that AI could take the load off. Maybe it can, for some of it. But a funder's application asks you to stand behind every statement in it, and a draft that came out of a tool doesn't change who is answering for it.
This article gives you a way to sort your grant process into tasks worth handing off and decisions you keep, plus a workflow that puts the human review in the right places.
Sort tasks by what happens when they're wrong
A useful test is not "can a tool do this?" but "what does it cost if this is wrong, and who notices?" A badly structured outline costs you ten minutes. A wrong eligibility call or an invented statistic can cost you the application, or worse.
Here is a starting sort. Adjust it to your own process.
| Task | Hand off a first pass | Keep for yourself |
|---|---|---|
| Reading a long notice and listing requirements | Yes | Confirm the list against the notice |
| Outlining the narrative to match the scoring criteria | Yes | Approve the structure |
| Drafting a first version from your own materials | Yes | Rewrite for accuracy and voice |
| Checking that sections match each other | Yes | Resolve conflicts |
| Deciding whether to apply | No | Yes |
| Choosing the strategy and the ask | No | Yes |
| Talking to the program officer and partners | No | Yes |
| Confirming every number, name and citation | No | Yes |
| Final read and sign-off | No | Yes |
Where handing off works
The tasks that suit a tool share a pattern: they start from material you already have, and you can check the result against that material.
- Reading and extracting. Long notices bury eligibility, scoring criteria, required attachments and deadlines in dozens of pages. A tool can pull out a first list. You still confirm it against the notice, since a missed requirement is an easy way to lose an application. How to read a funding notice covers what to look for.
- Outlining. Turning the review criteria into a skeleton for the narrative is mostly structure. From notice to outline walks through the method; a tool can produce a first outline for you to correct.
- First drafts. A blank page is slow. A draft built from your past proposals, reports and program descriptions gives you something to edit.
- Consistency checks. Does the budget match the narrative? Does the number of participants in the need statement match the one in the evaluation plan? Comparing sections is tedious and well suited to a second pair of eyes, human or not.
Where you stay in charge
Some work can't be handed off because it depends on things only you know.
- The go/no-go decision. Whether this grant fits your mission, capacity and calendar is a judgment about your organization. See the go/no-go decision.
- Relationships. A program officer's answer to a question, a partner's real commitment and a board member's support all come from conversations.
- Truth. Under 2 CFR 200.415(b), subrecipients certify to the pass-through entity, whenever they apply for funds, that the information they provide is "true, complete, and accurate" and that false information or the omission of a material fact may bring criminal, civil or administrative consequences. That rule covers subrecipients, but the principle is general: whoever signs for your organization is vouching for the text, whatever tool helped write it.
- Final review. Read the whole application as the reviewer will, in order, against the scoring table.
The National Science Foundation puts the same principle plainly in its notice on generative AI: proposers are responsible for the accuracy and authenticity of their submission, including content developed with AI tools.
Check the funder's rules before you start
Funders differ on AI, and some have written rules. The National Institutes of Health, for example, says applications that are substantially developed by AI, in whole or in a section, won't be considered the original ideas of the applicant. NSF encourages proposers to say in the project description whether and how they used generative AI. Funder AI policies goes through where to find these rules. Read them first, because they decide how much you can hand off at all.
What a draft depends on
The quality of what comes back depends on what goes in. A tool given only a notice and a one-line prompt writes in generalities, because it has nothing specific about your organization to work from. A tool given your last three proposals, your program descriptions and your outcome data can write something much closer to what you'd say.
That also means your input needs to be accurate. If an old proposal still lists a program you've ended or a staff member who left, a draft will repeat it with confidence. Before you hand material over, spend ten minutes removing what's out of date. Keep the list of what you shared, too, so that when you review the draft you know which files each claim should trace back to. Fact-checking AI drafts covers that routine in detail.
A workflow that keeps people in the loop
- Notice. Start from the funding notice itself, not from a blank prompt.
- Requirements list. Have a first pass extracted, then check it against the notice yourself.
- Concept. Decide the project idea and the angle for this funder. This is your call.
- Draft. Generate or write a first version from your organization's materials.
- Human edit. Rewrite for accuracy, specifics and voice. Cut anything you can't back up.
- Check against the notice. Go through the review criteria one at a time and mark each as covered, thin or missing.
- Submit. Someone with authority reviews and signs.
Say a rural food bank is applying for a state hunger-relief grant with a 12-page narrative. The development director hands off three jobs: extracting the 14 required items from the notice, outlining the narrative against the 5 scoring criteria, and drafting the organization background from the latest annual report. She keeps the decision to apply, the call to the program officer to ask whether pantry expansion counts as an eligible expense, the figure for households served, and the final read. The tool saves her the blank page. The parts that could sink the application stay with her.
Your next step
Write out every task in your last application, from first read of the notice to submission. Next to each, mark whether it needs your judgment, a relationship or a fact only you can confirm. Whatever is left unmarked is your candidate list for handing off.
SoliContraX is built for the stretch between "we should apply" and "submitted". Upload the notice, approve a concept, and work from a draft written from your own files and checked against what the funder asked for. Start your next application.